Re: ctime set by truncate even if NOCMTIME requested

[Trond Myklebust <trond.myklebust@fys.uio.no>]

må den 19.09.2005 Klokka 12:51 (-0500) skreiv Steve French:
> I am seeing requests to set ctime on truncate which does not make any 
> sense to me as I was testing with the flag that should have turned that 
> off.  ie my inodes having S_NOCMTIME set, as NFS does.
> 
> do_truncate (line 206 of open.c) sets
>       newattrs.ia_valid = ATTR_SIZE | ATTR_CTIME
> instead of
>        newattrs.ia_valid = ATTR_SIZE;
>       if(!IS_NOCMTIME(inode))
>            newattrs.ia_valid |= ATTR_CTIME;
> 
> I thought that the correct way to handle this for network filesystems, 
> is to let the server set the mtime and ctime unless the application 
> explicitly sets the attributes (in the case of the sys call truncate or 
> ftruncate the application is not explicitly setting the ctime/mtime as 
> it would on a backup/restore so they should be ignored for the network 
> fs so the server will set it correctl to its time, reducing traffic and 
> more accurately representing the time it got updated).
> 
> Shouldn't there be a IS_NOCMTIME check in the truncate path in fs/open.c?

See the discussion on this a couple of weeks back.

It is quite correct for the kernel to request that the filesystem set
ctime/mtime on successful calls to open(O_TRUNC).
  http://www.opengroup.org/onlinepubs/009695399/toc.htm

It is _incorrect_ for it to request that ctime/mtime be set (or that the
suid/sgid mode bit be cleared) if a truncate()/ftruncate() call results
in no size change.
  http://www.opengroup.org/onlinepubs/009695399/toc.htm

So the current do_truncate() does have to be changed. Adding a check for
IS_NOCMTIME would be wrong, though.

Cheers,
  Trond